JP7246272B2 - Ground subsidence prediction system - Google Patents

Description

The present invention relates to a land subsidence prediction system for predicting land subsidence.

2. Description of the Related Art Conventionally, for example, prediction of the state of the ground of a planned construction site for an unconstructed building has been performed. In the technique shown in Patent Document 1, first, the geological characteristic value of each layer is estimated from the boring data already obtained for a plurality of data around the estimated ground location, and a contour map of the stratum specific value is generated from the geological characteristic value. do. Next, the geological characteristic value of the estimated ground location is estimated from the generated contour map. A predetermined calculation formula is applied from the geological characteristic value of the estimated ground location and the specification value of the structure to be constructed there to calculate the value of the impact on the ground during construction of the structure.

Japanese Patent No. 5516211

However, in the ground subsidence prediction system shown in Patent Document 1, in order to create a contour map around an estimated ground location (estimated location for estimating ground subsidence), boring test data for a large number of locations around the estimated location is required. There is a need to. In addition, even if the state of the ground at the estimated point is known, it is not easy to accurately predict ground subsidence at the estimated point.

The present invention has been made in view of these points, and the present invention provides a ground subsidence prediction system capable of accurately predicting ground subsidence at a specific point as an estimated point of ground subsidence. offer.

In view of the above problems, the present invention provides a ground subsidence prediction system for predicting ground subsidence at a specific point as an estimated ground subsidence point, which is obtained by boring tests at construction points of a plurality of existing buildings. Ground information for learning including the depth from the ground surface of the construction site, the value of the soil parameter set by quantifying the soil quality according to the depth, and the N value in the standard penetration test according to the depth. , based on the amount of land subsidence of the existing building corresponding to the ground information for learning, and the value of the learning subsidence parameter set according to the likelihood of ground subsidence, as teacher data, the ground at an arbitrary point a first learning unit that learns the calculation of the subsidence parameter value of the arbitrary point from the information by machine learning; map data of a predetermined area; position information, the depth from the ground surface obtained by the boring test for each known point, the value of the soil parameter set by quantifying the soil according to the depth, and the N in the standard penetration test according to the depth an estimated point setting unit for setting the estimated point on the map; and a location around the estimated point set by the estimated point setting unit. A surrounding information extraction unit for extracting the known point information for each existing known point, and the known point information for each known point extracted by the surrounding information extraction unit are inputted to the first learning unit, and the known a peripheral subsidence parameter calculation unit that causes the first learning unit to calculate the value of the subsidence parameter for each point; and a subsidence parameter estimator that calculates the value of the subsidence parameter.

According to the present invention, in the first learning unit, for an existing building that has subsided so far, the ground information for learning of the construction site of the existing building and the value of the subsidence parameter for learning are used as teacher data, and an arbitrary point Calculation of the subsidence parameter value of the arbitrary point is learned by machine learning from the ground information. This machine learning may be performed by, for example, a deep neural network (DNN), which is one of the deep learnings described later, or may be performed using other algorithms.

Here, the construction sites targeted for this training data are the sites where land subsidence actually occurred. Area, values obtained by quantifying the specifications of the foundation of the building) are normalized by the quantified values of the learning subsidence parameters. Therefore, this subsidence parameter quantifies the likelihood of ground subsidence for buildings with the same specifications, in other words, it uses the high risk of ground subsidence as a variable. Since the larger the subsidence parameter, the easier it is for the ground to subside, the risk of ground subsidence is high, and for example, a foundation structure that is unlikely to subside is selected.

In the present invention, with a specific point as an estimated point of ground subsidence, map data of a predetermined area including the estimated point and known point information for each known point in a plurality of known points on the map of the map data are stored. An information storage unit is provided.

Here, when the estimated point is set by the estimated point setting unit, the surrounding information extraction unit extracts known point information for each known point existing around the estimated point. At this time, the peripheral information extraction unit may extract known points existing within a predetermined distance from the estimated point, or may extract a specific number of known points close to the estimated point.

Next, the surrounding subsidence parameter calculation unit inputs the known point information of these extracted known points to the first learning unit, and calculates subsidence parameter values of known points existing around the estimated point. This calculated subsidence parameter value is a value that affects the subsidence parameter value of the estimated point. Therefore, next, the subsidence parameter estimator calculates the value of the subsidence parameter of the estimated point based on the subsidence parameter of each known point existing around the estimated point.

In this way, in the present invention, the learning ground information at the construction site where subsidence actually occurred and the value of the learning subsidence parameter set based on the ground subsidence amount of the existing building that actually subsided are taught. Since learning is performed as data, the first learning unit can accurately learn the calculation of the subsidence parameter value at any point including the estimated point. Then, at the estimated point, the subsidence parameter value of the estimated point is calculated using the subsidence parameter values of the known points around it, so even if there is no ground information of the estimated point, for example, the position information of the estimated point is input. Only by doing so, even a designer with little experience can rationally and objectively predict the degree of easiness of subsidence at the estimated point with good accuracy. As a result, it is possible to prevent building failures due to land subsidence and reduce construction costs.

As a more preferable aspect, the surrounding information extraction unit extracts the known point information for each of the known points existing within a predetermined distance from the estimated point, and the ground subsidence prediction system extracts the surrounding information a test execution determination unit that determines whether or not to conduct a boring test and a standard penetration test at the estimated points based on the number of known points extracted by the extraction unit and the subsidence parameter values calculated by the subsidence parameter estimation unit; Prepare more.

Here, when extracting known point information for each known point existing within a predetermined distance from the estimated point, for example, if the number of extracted known points is less than a predetermined number, the calculated subsidence of the estimated point If the value of the parameter is greater than the predetermined value, or if both of them hold, there is a high possibility that the ground will subside at the estimated point. Therefore, according to this aspect, in the test implementation determination unit, considering such a case, it is possible to determine whether or not to implement the boring test and the standard penetration test at the estimated point, so the ground at the estimated point The possibility of subsidence can be predicted more accurately.

Furthermore, as a preferred embodiment, the ground subsidence prediction system comprises a learning subsidence parameter value of a selected point selected from a plurality of existing building construction points, and a plurality of surrounding construction points within a predetermined range from the selected point. and the learning positional information specifying the relative positional information of each of the surrounding construction sites with respect to the selected site, as one data group, and a plurality of data obtained for each of the different selected sites Using the data group of as teacher data, the value of the subsidence parameter of a plurality of peripheral points existing around an arbitrary point, and the relative position information of each peripheral point with respect to the arbitrary point, the arbitrary point The subsidence parameter estimation unit calculates the subsidence parameter value for each of the known points calculated by the surrounding subsidence parameter calculation unit, and Relative position information of each of the known points with respect to the estimated point is input to the second learning unit, and the second learning unit is caused to calculate a subsidence parameter value of the estimated point.

According to this aspect, the second learning unit learns the calculation of the subsidence parameter value of the predetermined construction site from the relative positional information of the surrounding construction sites and the subsidence parameter value thereof. Therefore, the subsidence parameter estimating section can cause the second learning section to calculate the value of the subsidence parameter of the estimated point, taking into account the directions of known points around the estimated point. This makes it possible to calculate a more accurate subsidence parameter value.

As a more preferable aspect, the second learning unit stores location information for each known point within the predetermined area stored in the area information storage unit, and the location information within the predetermined area calculated by the first learning unit. The value of the subsidence parameter for each known point of , was machine-learned as teacher data. According to this aspect, since the second learning unit learns the calculation of the subsidence parameter value in the predetermined area including the estimated point, it is possible to improve the accuracy of the subsidence parameter value of the estimated point.

In a more preferred embodiment, the subsidence parameter estimating unit, based on the value of the subsidence parameter for each of the known points calculated by the surrounding subsidence parameter calculating unit, and the distance from the estimated point to each known point extracted, A subsidence parameter value for the estimated point is calculated.

According to this aspect, the value of the subsidence parameter of the estimated point is greatly affected by the value of the subsidence parameter for each known point and the distance from the estimated point to each known point extracted. , the accuracy of the value of the subsidence parameter at the estimated point can be increased.

According to the present invention, it is possible to accurately predict ground subsidence at a specific point as an estimated ground subsidence point.

BRIEF DESCRIPTION OF THE DRAWINGS It is a schematic conceptual diagram for demonstrating the ground subsidence prediction system which concerns on 1st Embodiment of this invention. 2 is a block diagram of the ground subsidence prediction system shown in FIG. 1; FIG. FIG. 3 is a table showing teacher data for learning by a subsidence parameter learning unit (first learning unit) shown in FIG. 2; FIG. FIG. 3 is a schematic conceptual diagram of a neural network of a subsidence parameter learning unit (first learning unit) shown in FIG. 2; 3 is a diagram showing an example of map data stored in a storage unit shown in FIG. 2; FIG. 4 is a table showing an example of subsidence parameters for each known point calculated by a surrounding subsidence parameter calculator from ground information of each known point; 3 is a diagram for explaining a calculation method of a subsidence parameter estimator shown in FIG. 2; FIG. FIG. 3 is a flow chart of a series of processes by the ground subsidence prediction system shown in FIG. 2; It is a block diagram of a ground subsidence prediction system according to a second embodiment. FIG. 10 is a diagram showing an example of map data for explaining learning by a subsidence estimation learning unit (second learning unit) shown in FIG. 9; FIG. 10 is a schematic conceptual diagram of a neural network of a subsidence estimation learning unit shown in FIG. 9; FIG. 10 is a diagram for explaining a calculation method of a subsidence parameter estimator shown in FIG. 9; FIG. FIG. 10 is a flow diagram of a series of steps by the ground subsidence prediction system shown in FIG. 9;

Hereinafter, ground subsidence prediction systems according to first and second embodiments of the present invention will be described with reference to the drawings.

[First embodiment]
1. About the device configuration of the ground subsidence prediction system 1 The ground subsidence prediction system 1 according to the present embodiment uses, for example, a main server (host computer) 10 and a mobile terminal (communication terminal) 20 to predict a specific point of ground subsidence. As an estimated point, it predicts ground subsidence at the estimated point. The ground subsidence prediction system 1 will be described below.

The main server 10 of the land subsidence prediction system 1 includes an arithmetic device 10A having artificial intelligence that calculates (learns) the state of land subsidence (specifically, values of subsidence parameters), and inputs data to the arithmetic device 10A. and an input device 10B such as a keyboard. Arithmetic unit 10A stores data such as area information of a predetermined area to be described later, storage unit 10a composed of ROM or RAM storing programs machine-learned by artificial intelligence, and subsidence parameter values based on the machine-learned programs. and a computing unit 10b composed of a CPU or the like for computing . Specifically, the storage unit 10a stores the program of the subsidence parameter learning unit 11, the area information stored in the area information storage unit 12, and the like shown in FIG. The learning (calculation) by the subsidence parameter learning unit 11 shown is executed.

The mobile terminal 20 of the ground subsidence prediction system 1 is capable of communicating with the main server 10 via a network, and has an arithmetic device 20A for calculating subsidence parameter values at an estimated point, and a display such as a touch panel for inputting and displaying data. - An input unit 20B is provided. The arithmetic device 10A includes a storage unit 20a such as a ROM or RAM for storing data from the main server 10 and data input by the display/input unit 20B, and an arithmetic unit such as a CPU for calculating the subsidence parameter of the estimated point. 20b and . Specifically, the storage unit 20a stores data from the main server 10, data input by the display/input unit 20B, applications (programs) for calculation by the calculation unit 20b, and the like. , the peripheral information extractor 22, the peripheral subsidence parameter calculator 23, and the subsidence parameter estimator 24 shown in FIG.

In this embodiment, the ground subsidence prediction system 1 is provided with a main server 10 and a mobile terminal 20 separately. not to be

Further, the ground subsidence prediction system 1 may be configured as one system instead of being configured with the main server 10 and the mobile terminal 20 . Specifically, the storage unit 10a of the main server 10 stores data stored in the storage unit 20a of the mobile terminal 20, and the calculation unit 10b of the main server 10 performs calculation by the calculation unit 20b of the mobile terminal 20. Therefore, the mobile terminal 20 may be omitted. In addition, the mobile terminal 20 does not perform the above-described storage and calculation, and the main server 10 performs all the storage and calculation. It may serve only to output the value of the subsidence parameter of the estimated point. In this embodiment, software functions of the ground subsidence prediction system 1 will be described below with reference to the block diagram of FIG.

In this embodiment, the calculation device 10A of the main server 10 includes a subsidence parameter learning section (first learning section) 11 and a regional information storage section 12. FIG. The computing device 20A of the mobile terminal 20 includes an estimated point setting section 21, a surrounding information extracting section 22, a surrounding subsidence parameter calculating section 23, and a subsidence parameter estimating section . The data of the arithmetic unit 10A of the main server 10 and the data of the arithmetic unit 20A of the portable terminal 20 can be mutually transmitted and received via a network.

2. About the model of the ground subsidence prediction system 1 2-1. Subsidence parameter learning unit 11
The subsidence parameter learning unit 11 (hereinafter referred to as “first learning unit 11”) acquires ground information for learning at construction points P1, P2, P3, . . . of a plurality of existing buildings A, B, C, . , P2, P3, . It is learned by learning. Here, the "point" that is the target of the training data learned by the first learning unit 11 is a construction point that exists in the construction site where the existing building is constructed, and the ground subsidence actually occurs, and the amount of ground subsidence is measured. or a point within the construction site where the amount of land subsidence is estimated from the specifications of the existing building. The "point" that is the target of the data input to the first learning unit 11 after learning is a known point where a boring test and a standard penetration test have been performed in a predetermined area, and the data input from these test results The data output by is the value of the subsidence parameter at the known point.

Specifically, the "ground information for learning" is obtained by boring tests at the construction points P1, P2, P3, . . . of the existing buildings A, B, C, . It includes the depth from the ground surface, the value of the soil parameter set by quantifying the soil quality according to the depth, and the N value in the standard penetration test according to the depth. Here, the "depth" to be input is, for example, the depth that constitutes the soil, such as the depth of the bottom of the layer that constitutes each soil, and the "depth" that can specify the information of the layer thickness of the soil. The "depth" value associated with the soil parameter value and the N value is not particularly limited as long as it can be input.

Generally, in the boring test, the type of soil is investigated according to the depth from the ground surface. Here, the "soil parameter" is a parameter quantified and set for each soil property. For example, the soil quality can include clay, silt, fine sand, coarse sand, fine gravel, gravel, coarse gravel, cobble, or volta, and is quantified for each soil quality. For example, a numerical value corresponding to the particle size distribution according to the type of soil may be set. For example, as the particle size distribution according to the type of soil increases, the bearing capacity to support the load from the building increases (ground subsidence is less likely). . For example, as shown in FIG. 3, soil parameters A to E are assigned numeric values of soil parameters.

"N value in the standard penetration test" refers to the N value measured in the standard penetration test (test in accordance with JIS A 1219), the converted N value based on the Swedish sounding test (test in accordance with JIS A 1212), etc. The larger the N value, the greater the bearing capacity to support the load from the building.

As shown in FIG. 3, at each construction point P1, P2, P3, . Among these ground information for learning, the values of the soil parameters of each construction site P1, P2, P3, . can be set.

The "learning subsidence parameter" associated with each construction site is a variable that is set to a value according to the likelihood of ground subsidence based on the amount of ground subsidence of the existing building. It is set to a value that facilitates land subsidence.

In this embodiment, the value of the "learning subsidence parameter" is the amount of ground subsidence at the construction site, considering the specifications of the existing building (for example, the use of the building, the number of times, the building area, specification) is normalized by a numerical value. For example, this subsidence parameter quantifies the easiness of ground subsidence (risk of ground subsidence) for buildings with the same specifications. In other words, as the magnitude of the subsidence parameter increases, the ground subsidence becomes more likely, so a foundation structure that is less likely to subside is selected. Since these learning subsidence parameter values are teaching data that serve as the basis for learning, the amount of subsidence of the existing building and the specifications of the existing building can be used by experts to set the values based on their experience. Well, you can set it like this:

Specifically, the value of the "learning subsidence parameter" may be a value obtained by dividing the ground subsidence amount of the existing building by a value corresponding to the building specifications of the existing building shown below. In the present embodiment, the "value corresponding to the building specifications of the existing building" is, for example, (1) a usage parameter value set by quantifying the usage of the existing building, (2) the number of floors of the existing building, and (3) It is a value obtained by multiplying the building area of the existing building and (4) the value of the foundation parameter set by digitizing the foundation type selected for the existing building.

"Usage parameter" is a parameter according to the usage of the existing building. is digitized and set according to For example, the application parameter values are assigned numerical values in descending order of the degree of land subsidence.

The "building floor number" is the number of floors of an existing building or a non-constructed building, and the "building area" is an area corresponding to the base portion of the existing building or non-constructing building.

"Foundation parameters" are parameters set by digitizing the foundation type selected for an existing building or an unconstructed building. Specifically, (1) spread foundation structure with piles, (2) pile draft foundation structure with piles, (3) pile foundation structure with friction piles, and (4) support piles supported by support layers. The pile foundation structure etc. can be mentioned, and these foundation types (form of foundation structure) are set numerically. For example, numerical values are assigned in descending order of the subsidence of the foundation structure.

Furthermore, in the case of a pile foundation structure, depending on the length of the friction pile, the type of the friction pile, etc., the numerical value of the size may be set according to the order of susceptibility to ground subsidence. Furthermore, in addition to the foundation structure, for example, a foundation structure improved with cement milk may also be included, and the above-mentioned numerical values are set for this structure as well.

In this way, as shown in FIG. 3, the value of the "learning subsidence parameter" for each construction site is set as training data in association with the learning ground information for each construction site. For example, the value of the learning subsidence parameter is 50 at the construction point P1, and the value of 32 is the learning subsidence parameter at the construction point P2. Therefore, if buildings of the same specifications are constructed at the construction point P1 and the construction point P2, the construction point P1 is more likely to undergo land subsidence. The construction points P1 to P4 shown in FIG. 3 have ground that is likely to subside in the order of construction points P4, P1, P3, and P2.

Using the ground information for learning and the value of the settlement parameter for learning for each construction site P1, P2, . . . shown in FIG. Learn by machine learning.

In this embodiment, the first learning unit 11 determines that the output values calculated with respect to the input values of the existing buildings A, B, C, . Then, based on the actual amount of land subsidence, machine learning is used to learn the calculation of the subsidence parameter value so as to converge to the value of the subsidence parameter (actual subsidence parameter) set according to the likelihood of land subsidence. .

This learning may be performed, for example, by repeatedly correcting a correction coefficient by which each variable is multiplied to a predetermined mathematical formula consisting of variables corresponding to the number of input values of the ground information for learning. In this embodiment, as shown in FIG. 4, the first learning unit 11 is provided with a deep neural network (DNN: hereinafter referred to as "neural network") 11'. is the input value, and the value of the subsidence parameter is calculated as the output value.

In this embodiment, the neural network 11' has an input layer 11A. The input layer 11A receives depths N1, N2, . The input layer 11A is composed of a plurality of input layer neuron elements 11a corresponding to the number of input values. The number of input layer neuron elements 11a in the input layer 11A can be increased or decreased according to the number of input condition data, and a model of the neural network 11' is constructed according to the number of input data due to this increase or decrease.

The neural network 11' has an output layer 11E. The output layer 11E is composed of one output layer neuron element 11e that outputs the value of the subsidence parameter. The neural network 11' has three intermediate layers 11B, 11C, 11D. Three intermediate layers 11B, 11C, 11D are provided between the input layer 11A and the output layer 11E. Each intermediate layer 11B, 11C, 11D includes a plurality of intermediate layer neuron elements 11b, 11c, 11d directly or indirectly coupled to these elements.

In this embodiment, the intermediate layer is composed of three layers, but for example, the intermediate layer may be composed of one, two, four or more layers, It is not limited to layers. Further, the number of intermediate layer neuron elements 11b, 11c, and 11d constituting each of the intermediate layers 11B, 11C, and 11D corresponds to the number of input layer neuron elements 11a. The number of neuron elements 11a may be changed, and each intermediate layer may include the number of intermediate layer neuron elements different from the number of input data.

The intermediate layers 11B, 11C, and 11D perform predetermined calculations using the neuron parameter values of the neuron elements in the same layer on the input layer 11A side, and output the calculation results to the neuron elements on the output layer 11E side. It is. Specifically, the intermediate layer neuron elements 11b, 11c, and 11d and the output layer neuron element 11e receive the neuron parameter values input from the input layer neuron element 11a or the intermediate layer neuron elements 11b, 11c, and 11d on the input layer 11A side. is used to perform a predetermined calculation.

Specifically, each of the intermediate layer neuron elements 11b, 11c, 11d and the output layer neuron element 11e, which execute operations, has a predetermined activation function, and the input data (parameter values) are Calculate the value of the neuron parameter by substituting it into the activation function. For example, each intermediate layer neuron element 11b of the intermediate layer 11B receives the input value of each input layer neuron element 11a as a neuron parameter value, and the activation function calculates the neuron parameter value for each input layer neuron element 11a. be done.

At this time, the values of the neuron parameters output from the intermediate layer neuron elements 11b, 11c, and 11d of the intermediate layers 11B, 11C, and 11D correspond to the neuron parameter values calculated by the activation function in the elements. , is multiplied by a weighting factor, and is output to the output layer neuron element 11e or the intermediate layer neuron elements 11c and 11d on the output layer 11E side.

In this embodiment, the subsidence parameter values of the existing buildings A, B, C, . The weighting coefficients multiplied by the neuron parameter values of the intermediate layer neuron elements 11b, 11c, and 11d are repeatedly corrected until convergence within the set range. In this manner, in this embodiment, the calculation (method) of the subsidence parameter value can be learned in advance by machine learning. In this way, the machine learning of the first learning unit 11 is executed using the ground information for learning and the values of the subsidence parameters for learning at a plurality of construction points P1, P2, P3, . . . From the ground information, a model is generated for calculating subsidence parameter values.

In the present embodiment, the weighting coefficients multiplied by the neuron parameter values calculated by the activation functions in the intermediate layer neuron elements 11b, 11c, and 11d are assigned to the intermediate layer neuron elements 11b, 11c, and 11d. In addition to this, the output layer neuron element 11e may also be provided with a weighting factor by which the value of the neuron parameter calculated by the activation function is multiplied.

2-2. Regional information storage unit 12
As shown in FIG. 5, the area information storage unit 12 stores map data of a predetermined area and known point information of a plurality of known points T1, T2, . . . on the map M1 of the map data. This known point information is set by digitizing the position information of each of the known points T1, T2, ..., the depth from the ground surface obtained by the boring test of each of the known points T1, T2, ..., and the soil quality corresponding to the depth. These data include the values of the soil parameters and the N values in the standard penetration test according to depth, and these data are linked to each known point. Since the ground information of depth, soil parameter value, and N value shown here are the same types of information as those explained in the ground information for learning, detailed explanation will be omitted. Note that the known points on the map M1 of the map data may include the construction points used in the machine learning.

Specifically, each known point is a point where a boring test and a standard penetration test were performed, and the ground information obtained by these tests is the position information of each known point T1, T2, ... on the map M1. tied to. It should be noted that, at this known point, it is not necessary that a building has been constructed at present or in the past. A subsidence parameter associated with the estimated or estimated ground subsidence amount may also be stored. Note that the predetermined area that is the target of the map data may be, for example, an area such as a ward or city, or an area that includes one or a plurality of common soil layers continuously. Such map data is transmitted from the main server 10 to the mobile terminal 20 and displayed on the display/input unit 20B of the mobile terminal 20 .

2-3. Estimated location setting unit 21
As shown in FIG. 5, the estimated point setting unit 21 sets an estimated point G on the map M1. In this embodiment, the operator inputs the position information of the estimated point G on the map M1 displayed on the display/input unit 20B of the mobile terminal 20 . As a result, the estimated point G on the map M1 is set in the arithmetic unit 20A of the mobile terminal 20. FIG.

2-4. Peripheral information extraction unit 22
As shown in FIGS. 2 and 5, the surrounding information extraction unit 22 extracts known point information for each known point T1, T7, T8, . (see Figure 6). In this embodiment, as shown in FIGS. 5 and 7, known point information is extracted for each known point T1, T7, T8, .

Here, in the present embodiment, the surrounding information extraction unit 22 extracts known point information for each known point existing within a predetermined distance r from the estimated point G. FIG. Specifically, as shown in FIGS. 5 and 7, the area extracted by the peripheral information extraction unit 22 is a circular area (a circle with a radius of 3 km) within a certain distance r (for example, 3 km) from the estimated point G. is. However, the peripheral information extraction unit 22 may extract known point information of a specific (constant) number of known points in ascending order of distance from the estimated point G, for example.

2-5. Surrounding subsidence parameter calculator 23
The surrounding subsidence parameter calculator 23 inputs the known point information for each of the known points T1, T7, T8, . . . The first learning unit 11 is caused to calculate the value of the subsidence parameter for each of the points T1 and T7. As described above, since the first learning unit 11 has learned to calculate the value of the subsidence parameter for an arbitrary point from the ground information of the arbitrary point, the subsidence parameter for each of the known points T1, T7, T8, . . . can be calculated with high accuracy.

Here, if the area information storage unit 12 already stores the actual subsidence parameter value for the extracted known point, the surrounding subsidence parameter calculation unit 23 calculates the subsidence parameter value for the known point. It is not necessary to calculate the value. As another aspect, even if the actual subsidence parameter values that are not calculated are already stored, the peripheral subsidence parameter calculation unit 23 can calculate the subsidence parameter values at the known points. is calculated, and the subsidence parameter value calculated at another known point may be corrected based on the difference or ratio between the calculated subsidence parameter value and the actual subsidence parameter value.

2-6. Subsidence parameter estimation unit 24
As shown in FIG. 7, the subsidence parameter estimator 24 calculates the value of the subsidence parameter for the estimated point G based on the subsidence parameter for each of the known points T1, T7, T8, etc. calculated by the surrounding subsidence parameter calculator 23. do. Here, the value of the subsidence parameter of the estimated point G may be calculated by, for example, the average value of the subsidence parameter calculated for each of the known points T1, T7, T8, .

However, in the present embodiment, the subsidence parameter value for each of the known points T1, T7, T8, . . . . . , the value of the subsidence parameter of the estimated point G is calculated. As a result, the subsidence parameter values of the estimated point G are the subsidence parameter values of the known points T1, T7, T8, . , L7, L8, .

More specifically, in this embodiment, the subsidence parameter estimating unit 24 sets the subsidence parameter so that the value of the subsidence parameter of the estimated point G decreases as the distance of each extracted known point from the estimated point increases. After correcting the value of , the average value of these values is estimated as the value of the subsidence parameter at the estimated point G. Thereby, the value of the subsidence parameter of the known point closer to the estimated point G can be reflected in the value of the subsidence parameter of the estimated point G. Such calculation may be performed based on Equation 1 shown below.

Here, DG is the value of the subsidence parameter of the estimated point, Dk is the value of the subsidence parameter of each extracted known point, Lk is the distance of each extracted known point from the estimated point, r is the distance for extracting known points from the estimated point, and N is the number of extracted known points. Note that α is a correction coefficient, and this correction coefficient is a constant set so that, for example, the value of the subsidence parameter at the estimated point calculated by Equation 1 matches the actual value of the subsidence parameter at the estimated point. is. However, this formula does not include Dk of known points that have not been extracted by the surrounding information extraction unit 22, and the known points that have not been extracted are not counted in the number of N.
Therefore, applying this Equation 1 to the case shown in FIG. 7 results in Equation 2 below.

If actual subsidence parameter values are already stored for some of the extracted known points in the area information storage unit 12, the subsidence parameter estimation unit 24 calculates the subsidence parameter Actual subsidence parameters may be used to calculate the value of .

3. Land Subsidence Prediction Method Using Land Subsidence Prediction System 1 A land subsidence prediction method using the land subsidence prediction system 1 according to the present embodiment will be described below with reference to FIG. First, in step S1, the first learning unit 11 performs machine learning using ground information for learning and subsidence parameter values for learning. As a result, a model of the neural network 11' that outputs subsidence parameter values from the ground information is constructed.

Next, in step S2, map data of a predetermined area and known point information of a plurality of known points T1, T2, . . . Note that the steps S1 and S2 are executed by the main server 10 regardless of the order. Next, in step S3, the application of the portable terminal 20 is started, and the map M of the map data stored in the main server 10 is displayed together with the known points T1, T2, .

Next, in step S<b>4 , the operator inputs the positional information of the estimated point G on the map M displayed on the mobile terminal 20 . Thereby, the estimated point setting unit 21 sets the estimated point G on the map M by linking the position information of the estimated point G to the map data. Next, in step S5, the surrounding information extraction unit 22 extracts known point information (around information) for each known point T1, T7, T8, .

Next, in step S6, the peripheral subsidence parameter calculator 23 transmits the ground information for each of the known points T1, T7, T8, . The first learning unit 11 in the server 10 calculates subsidence parameter values D1, D7, D8, . . . for each known point T1, T7, T8, . The first learning unit 11 transmits the calculated result to the arithmetic device 20A of the mobile terminal 20 via the network, and the result is input to the subsidence parameter estimating unit 24 .

Next, in step S7, the subsidence parameter estimator 24 calculates distances L1, L7, L8, . In step S8, the subsidence parameter estimating unit 24 generates subsidence parameter values D1, D7, D8, . . . for each known point T1, T7, T8, . The subsidence parameter value of the estimated point G is calculated based on L8. Finally, in step S9, the subsidence parameter value of the estimated point G is input to the display/input unit 20B of the mobile terminal 20, and the result is displayed.

In this embodiment, the learning ground information at the construction site where subsidence actually occurred and the value of the learning subsidence parameter set based on the actual subsidence amount of the existing building are learned as teacher data. , the first learning unit 11 can accurately learn to calculate the value of the subsidence parameter at an arbitrary point. At the estimated point G, the subsidence parameter value of the estimated point G is calculated using the subsidence parameter values of the known points in the surrounding area. The degree of easiness can be accurately predicted.

[Second embodiment]
A ground subsidence prediction system 1 according to a second embodiment of the present invention will be described below with reference to FIGS. 9 to 13. FIG. The ground subsidence prediction system 1 according to the second embodiment differs from that of the first embodiment in that: (1) As shown in FIG. (2) newly provided test execution determination unit 25; and the corrected (re-learned) points. Therefore, the detailed description of the same configuration as in the first embodiment will be omitted, and the different configuration will be described below.

The subsidence estimation learning unit 13 (hereinafter referred to as the “second learning unit 13”) calculates subsidence parameter values of a plurality of peripheral points existing around an arbitrary point and the relative position of each peripheral point with respect to the arbitrary point. Based on the information, the calculation of the subsidence parameter value at an arbitrary point is learned by machine learning.

For example, as shown in FIG. 10, the training data of the second learning unit 13 includes the values of the learning subsidence parameters of selected points selected from the construction sites W1 to W37 of a plurality of existing buildings, and the A set of training data (one data group ), which are multiple data groups obtained at different selected points.

For example, in FIG. 10, the construction point W1 on the map M2 of the predetermined area is selected as the selected point. Peripheral construction points W6, W7, W9, . . . within a predetermined range R1 (for example, distance r) from the selected construction point W1 are extracted. Then, the subsidence parameter value of the construction site W1 and the subsidence parameter values of each of the surrounding construction sites W6, W7, W9, . . . are used as learning subsidence parameter values. It is used as learning position information specifying relative positions of surrounding construction sites W6, W7, W9, . . . with respect to the construction site (selected site) W1. The learning position information of the surrounding construction sites W6, W7, W9, . The construction sites used in the learning of the second learning section may include the construction sites used in the learning of the first learning section.

Using these as a set of training data, as shown in FIG. The positional information for construction is input to the neural network 13', and the calculation of the subsidence parameter value is learned by machine learning so that the output value converges to the value of the subsidence parameter for learning of the construction site W1. By selecting such learning as a selection point for a plurality of other construction points (for example, W4, W36, etc. in FIG. 10), teacher data is extracted, and this teacher data is used to calculate settlement parameters. Learning to calculate values by machine learning.

As with the neural network 11' of the first learning unit 11, the neural network 13' of the second learning unit 13 also has an input layer 13A, intermediate layers 13B to 13D, and an output layer 13E. These layers are composed of neuron elements 13a to 13e, and perform machine learning by repeatedly correcting weighting coefficients multiplied by neuron parameter values.

Here, for example, the subsidence parameter values used in the training data at each of the construction sites W1 to W34 may be the subsidence parameter values of each construction site calculated by the first learning unit 11. It may be the value of a subsidence parameter. Here, it is preferable that the construction point storing the actual subsidence parameter value among the teaching data is selected as the selected point. As a result, the machine learning of the second learning unit 13 is performed so as to converge to the actual subsidence parameter value, so that the subsidence parameter value can be calculated with higher accuracy.

Furthermore, in the present embodiment, the second learning unit 13 extracts seven construction sites W6, W7, W9, . However, by selecting other construction points or changing the size of the predetermined range, the surrounding construction points can be , may perform similar learning. As a result, the value of the subsidence parameter of the estimated point G can be more accurately calculated by the second learning section 13 described later according to the number of known points extracted by the surrounding information extraction section 22 .

Here, the second learning unit 13 uses position information for each of the known points T1 to T33 in the predetermined area stored in the area information storage unit 12 and the known points calculated by the first learning unit 11 as teacher data. It may be machine-learned using subsidence parameter values for each of T1 to T33. As a result, the second learning unit 13 learns the calculation of the subsidence parameter value in the predetermined area including the estimated point G, so that the accuracy of the subsidence parameter value of the estimated point G can be improved.

In this embodiment, as shown in FIG. 12, the subsidence parameter estimation unit 24 calculates subsidence parameter values D1, D7, T8, . and relative positional information (X1, Y1), (X7, Y7), (X8, Y8), . to calculate the value of the subsidence parameter of the estimated point G.

Here, as shown in FIG. 12, the position information for each of the known points T1, T7, T8, . Coordinates in a polar coordinate system with G as the origin may be used.

In the second learning unit 13, for the selected construction site W1, a neural network model generated according to the number of construction sites W6, W7, W9, . . . Select and use this model to calculate the value of the subsidence parameter for the estimated point.

Based on the number of known points extracted by the surrounding information extraction unit 22 and the subsidence parameter value calculated by the subsidence parameter estimation unit 24, the test execution determination unit 25 determines whether or not the boring test and the standard penetration test should be performed at the estimated points. judge. Specifically, when the number of known points extracted by the surrounding information extraction unit 22 is less than a predetermined number (for example, less than 6), the number of points extracted is small. , boring test and standard penetration test at the estimated point G are necessary.

On the other hand, if the number of known points extracted by the surrounding information extraction unit 22 is equal to or greater than a predetermined number (for example, 6 or more), the number of extracted points is sufficient. The boring test and the standard penetration test at point G will be rejected. For example, in this embodiment, the number of known points is seven, so it is determined that the test should not be performed.

Furthermore, even if it is determined that the test should not be performed, if the subsidence parameter value calculated by the subsidence parameter estimation unit 24 exceeds a predetermined value (for example, exceeds 30), the ground Since the possibility of subsidence is high, the test implementation determination unit 25 determines that the boring test and the standard penetration test at the estimated point G should be performed. On the other hand, if the subsidence parameter value calculated by the subsidence parameter estimation unit 24 is equal to or less than a predetermined value (for example, 30 or less), the possibility of ground subsidence is low. It is determined that the boring test and standard penetration test are not performed.

In this way, when the number of known points extracted by the surrounding information extraction unit 22 is less than a predetermined number, or when the subsidence parameter value calculated by the subsidence parameter estimation unit 24 exceeds a predetermined value In addition, the test execution determination unit 25 determines that the boring test and the standard penetration test at the estimated point G are required. As a result, the possibility of ground subsidence at the estimated point G can be predicted more accurately.

A ground subsidence prediction method using the ground subsidence prediction system 1 according to this embodiment will be described below with reference to FIG. In steps S1 and S2, the same operations as steps S1 and S2 shown in FIG. 8 are performed. In this embodiment, in step S21, for the first learning unit 11 that performed machine learning in step S1, among the data of known points stored in step S2, for known points where ground subsidence actually occurred, Using the ground information and the subsidence parameter as teacher data, learning by the first learning unit 11 is performed again. As a result, the model generated in step S1 is corrected, and the first learning unit 11 can re-learn the tendency of land subsidence in the predetermined area.

In steps S3 to S6, the same operations as steps S3 to S6 shown in FIG. 8 are performed. Next, in step S61, the subsidence parameter estimator 24 calculates coordinates of known points as shown in FIG.

In step S81, the subsidence parameter estimator 24 transmits position information (X1, Y1), (X7, Y7), (X8, Y8) for each known point T1, T7, T8, . ), and the associated subsidence parameter values D1, D7, D8, . . . The second learning unit 13 in the main server 10 calculates the subsidence parameter value of the estimated point G. FIG. The second learning unit 13 transmits the calculated result to the arithmetic device 20A of the mobile terminal 20 via the network, and the result is input to the test execution determination unit 25 .

Next, in step S82, the test execution determination unit 25 determines whether or not the boring test and the standard penetration test of the estimated point G are to be performed based on the number of extracted known points and the calculated subsidence parameter value of the estimated point G. judge. Finally, in step S9, the value of the subsidence parameter of the estimated point G and the result of necessity of test implementation are input to the display/input unit 20B of the portable terminal 20, and the result is displayed.

Although the embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and various designs can be made without departing from the spirit of the invention described in the claims. Changes can be made.

In the second embodiment, the test execution determination unit determines whether or not to conduct a boring test and a standard penetration test at the estimated point based on the number of extracted known points and the calculated subsidence parameter value at the estimated point. For example, the test implementation determination unit uses one or more selected from the average value of the distance from the estimated point to the known point extracted, the number of extracted known points, or the calculated subsidence parameter value of the estimated point, It may be determined whether or not a boring test and a standard penetration test should be conducted at the estimated point. Furthermore, the first learning unit may be used to create a map indicating values of subsidence parameters linked to map data in a predetermined area.

1: Ground subsidence prediction system 10: Main server 20: Mobile terminal 11: Subsidence parameter learning unit (first learning unit) 12: Regional information storage unit 13: Subsidence estimation learning unit (second learning unit) 21: Estimated point setting unit, 22: Surrounding information extractor, 23: Surrounding subsidence parameter calculator, 24: Subsidence parameter estimator, 25: Test implementation determination unit, G: Estimated point, T1 to T33: Known points

Claims (5)

A ground subsidence prediction system that predicts ground subsidence at a specific point as an estimated point of ground subsidence,
The depth from the ground surface of the construction site, the value of the soil parameter set by quantifying the soil quality according to the depth, and the standard intrusion according to the depth, obtained by boring tests at the construction site of a plurality of existing buildings. A learning subsidence parameter value set according to the likelihood of ground subsidence based on the ground subsidence information for learning including the N value in the test and the amount of land subsidence of the existing building corresponding to the ground information for learning. and a first learning unit that learns by machine learning the calculation of the subsidence parameter value of the arbitrary point from the ground information of the arbitrary point, using as teacher data,
Map data of a predetermined area, location information for each known point at a plurality of known points on the map of the map data, depth from the ground surface obtained by a boring test for each known point, and depending on the depth a region information storage unit that stores known point information including a soil parameter value set by quantifying the soil quality and the N value in the standard penetration test according to the depth;
an estimated point setting unit that sets the estimated point on the map;
a peripheral information extraction unit for extracting the known point information for each of the known points existing around the estimated point set by the estimated point setting unit;
The known point information for each of the known points extracted by the surrounding information extraction unit is input to the first learning unit, and the first learning unit calculates the value of the subsidence parameter for each known point. a calculation unit;
a subsidence parameter estimator that calculates the value of the subsidence parameter of the estimated point based on the subsidence parameter for each of the known points calculated by the surrounding subsidence parameter calculator. . The peripheral information extraction unit extracts the known point information for each of the known points existing within a predetermined distance from the estimated point,
The ground subsidence prediction system uses the number of known points extracted by the surrounding information extraction unit and the subsidence parameter values calculated by the subsidence parameter estimation unit to perform a boring test and a standard penetration test on the estimated points. 2. The ground subsidence prediction system according to claim 1, further comprising a test execution determination unit that determines necessity. The ground subsidence prediction system provides a learning subsidence parameter value of a selected point selected from a plurality of existing building construction sites, and a learning subsidence parameter value of a plurality of surrounding construction points within a predetermined range from the selected point. and positional information for learning that specifies the relative positional information of each of the construction sites in the surrounding area with respect to the selected site as one data group, and a plurality of data groups obtained for each of the different selected sites as teacher data. Calculating the value of the subsidence parameter of the arbitrary point from the values of the subsidence parameters of a plurality of surrounding points existing around the arbitrary point and the relative position information of each of the surrounding points with respect to the arbitrary point further comprising a second learning unit learned by machine learning,
The subsidence parameter estimating unit calculates the value of the subsidence parameter for each of the known points calculated by the surrounding subsidence parameter calculating unit, and relative positional information of each known point with respect to the estimated point, to the second learning unit. 3. The land subsidence prediction system according to claim 1 or 2, wherein the subsidence parameter value of the estimated point is calculated by the second learning unit. The second learning unit stores position information for each known point in the predetermined area stored in the area information storage unit, and subsidence information for each known point in the predetermined area calculated by the first learning unit. 4. The ground subsidence prediction system according to claim 3, wherein the parameter values are machine-learned as teaching data. The subsidence parameter estimating unit calculates the subsidence parameter of the estimated point based on the value of the subsidence parameter for each of the known points calculated by the surrounding subsidence parameter calculating unit and the distance to each known point extracted from the estimated point. 3. The ground subsidence prediction system according to claim 1 or 2, wherein the value of is calculated.

Images